Updated October 24, 2026. Checked against vendor documentation and launch coverage.
Search for “secret AI video commands” and you will find endless lists promising a hidden code that unlocks cinematic output. Most of them are wrong in a structural way: they describe a control panel that does not exist.
Here is what actually exists. Google Flow has…
![AI Shortcuts, Slash Commands and Camera Control: The Master Prompt Workflow for AI Video and Images (2026) Updated October 24, 2026 · Checked against vendor documentation and launch coverage Search for “secret AI video commands” and you will find endless lists promising a hidden code that unlocks cinematic output. Most of them are wrong in a structural way: they describe a control panel that does not exist. Here is what actually exists. Google Flow has no slash-command console. Veo and Kling do not parse a secret grammar. Midjourney does have a literal grammar (parameters), and Claude Code has real slash commands that you can write yourself. Everything else sold as a “shortcut” is a well-built template, and a template is only as good as the director’s thinking behind it. There is a second problem: most guides ranking today are stale. If your shortlist still includes Sora, you are working from an outdated map. OpenAI closed the Sora app on April 26, 2026 and switched off the Sora API on September 24, 2026. This guide separates what is real from what is recycled. You will get a reusable shot brief that works across video tools, camera and lighting vocabulary that models actually respond to, current controls for Google Flow, Kling 4.0 and Midjourney, and a practical way to turn Claude, Claude Code and Ollama into your own command layer. Three Things People Mean by “AI Shortcut” The phrase covers three very different mechanisms. Mixing them up is the root of most bad advice. Layer Where it lives Example Reliability True slash commands Claude Code, the interactive Ollama session /shotlist, /show info Deterministic: the software parses them Parameters Midjourney --ar 9:16, --stylize 150 Deterministic, but version-specific Structured prompt templates Veo and Flow, Kling, Claude chat Subject, action, camera, light, audio Probabilistic: the model interprets them A command modifier is a real shortcut only when software parses it the same way every time. Everything in the third row is a checklist that forces you to say what you mean. That is still valuable, because vague prompts, not weak models, are the most common reason for flat footage. What Changed in 2026: A Fact Check Before you copy any prompt library, verify it against this timeline. • Sora has been discontinued. OpenAI announced the wind-down in March 2026, closed the app on April 26 and the API on September 24. OpenAI support reportedly had no replacement product to point users to as of early October. A report on October 12 that OpenAI is in talks with a filmmaker-focused startup about Sora access is unconfirmed, so do not build a workflow on it. • Google Flow became an image-and-video studio. Veo 3.1 arrived on October 15, 2025 with Ingredients to Video, Frames to Video, Extend and Insert. In January 2026 Google added native vertical 9:16 output and 1080p and 4K upscaling. At I/O on May 19, 2026 it introduced Gemini Omni Flash, Flow Agent, Flow Tools and dedicated mobile apps. • Kling 4.0 launched on October 15, 2026, after an early-access Flash build opened on September 28. Headline specs: 30-second single-pass clips, up to 10 keyframes and up to 15 reference inputs. • Claude Code merged custom commands into skills. A file at .claude/commands/review.md and a skill at .claude/skills/review/SKILL.md both create /review. • Midjourney moved to V8.x. Parameter support shifted again, and some older advice (such as --q and --turbo) no longer applies to current versions. One of those facts should change how you plan this quarter: if a tutorial teaches Sora prompting as a live skill, treat the rest of it with suspicion. The craft transfers; the tool does not. The Master Prompt Workflow: A Six-Layer Shot Brief Google’s own Veo prompting guide breaks a prompt into subject, action, scene or context, camera angle, camera movement, composition and ambiance, with style layered on top. OpenAI’s now-retired Sora guide pushed the same idea from another direction: brief the model the way you would brief a cinematographer. Two vendors, one lesson. Specific, separable instructions beat poetic paragraphs. Use this brief for every shot, in any tool. Layer Question it answers Example language 1. Subject Who or what is on screen? A tired baker in her fifties, flour-dusted navy apron 2. Action What single thing happens? Slides a tray of loaves out of the oven 3. Camera Where is the lens and how does it move? Medium shot, eye level, slow dolly-in 4. Lens and focus How much of the frame is sharp? 50mm feel, shallow depth of field, focus on her hands 5. Light Source, direction, quality, colour Warm oven glow from camera right, cool window fill from left 6. Look and audio Grade, palette, sound Muted teal and amber, soft ambient hum, a single clink of metal The template, ready to paste and adapt: SHOT [number], [duration] seconds, [aspect ratio] SUBJECT: [who/what, 2-3 identifying details] ACTION: [one verb phrase, one beat] CONTEXT: [where, when, weather] CAMERA: [shot size] at [angle], [one movement], [speed] LENS AND FOCUS: [focal length feel], [depth of field], [focus target] LIGHT: [key source + direction + quality + colour temperature] LOOK: [grade or film stock, 3-5 colour anchors] AUDIO: [ambient, effects, one line of dialogue in quotes] AVOID: [extra people, on-screen text, jump cuts] Three rules make the template work. One camera move and one light setup per shot. Generators drift when a prompt asks for a whip pan, a dolly and a rack focus inside six seconds. If you want complexity, cut to a second shot. Repeat descriptors word for word. Character consistency starts with identical wording across shots, then gets reinforced by reference images. Say “static” when you mean static. Models tend to add subtle motion on their own. A phrase like “locked-off tripod shot, no camera movement” removes the guesswork. Camera Angles and Movements That Models Understand Move What it does Phrase to use Dolly in / out Pushes toward or away from the subject Slow dolly-in on her face Tracking Follows the subject sideways or from behind Tracking shot alongside the cyclist at hip height Pan / tilt Rotates horizontally or vertically Slow pan left across the skyline Crane or jib Rises or descends Crane up from the street to rooftop level Orbit Circles the subject Orbit 90 degrees clockwise around the watch Handheld Adds documentary wobble Handheld, slight shake, close on the chef Low / high angle Changes power dynamics Low-angle shot looking up at the statue Rack focus Shifts the sharp plane Rack focus from the glass to her eyes A caution from Google’s documentation: some advanced camera angles are not officially supported. When a phrase fails twice, simplify it instead of adding adjectives. Lighting Vocabulary That Changes the Mood • Golden hour: low warm sun, long shadows, friendly mood. • Rembrandt lighting: a key light from one side that leaves a small triangle of light on the shadowed cheek. • Rim or backlight: a bright edge that separates the subject from the background. • Soft window light: diffused daylight, gentle contrast, natural portraits. • Low-key: deep shadows with one controlled highlight, good for tension. • Volumetric light: visible beams through haze or dust. • Neon and practicals: coloured light from signs and lamps that appear in the frame. Always name the direction, such as “from camera left.” Light without a direction produces flat results. How to Use AI for Generating Videos: A Six-Step Workflow The fastest way to waste credits is to type a long prompt and hope. Professionals work in stages. 1. Write the intent in one sentence. Audience, platform, emotion. “A 9:16 product teaser that feels calm and premium.” 2. Break it into shots. Four to six shots of five to eight seconds usually beat one long generation. 3. Generate keyframes as still images. Midjourney or Flow’s image tools let you lock composition and light before you spend video credits. 4. Animate with image to video. Feed the still in, then describe only the motion and sound. 5. Extend, edit or stitch. Flow’s Extend and Insert, or Kling’s keyframe and editing tools, carry the shot forward. 6. Grade and assemble. Finish in your editor. Check the audio, the hands and any text before publishing. Why Image to Video Gives You More Control A still image already decides subject, framing, palette and light. The video prompt can then focus on one thing: movement. Google’s Frames to Video lets you set a start image and an end image, and Flow generates the bridge between them. Kling 4.0 expands that to as many as ten keyframes placed anywhere on the timeline. Think of keyframes as waypoints. You decide where the shot must be at second 0, 12 and 28, and the model fills in the gaps. When you write an image-to-video prompt, do not redescribe the picture. Describe what changes: “Slow push-in. Steam rises from the cup. Rain begins on the window. Soft piano under distant traffic.” Google Flow: The Commands That Actually Exist The search term “Google Flow commands” has a plain answer: there is no slash syntax. What Flow gives you instead is a set of named tools. • Frames to Video: supply a first and last image. • Ingredients to Video: use reference images, up to three in the original Veo 3.1 release, to steer characters, objects and style. • Extend: continue a clip from its final second, which is how creators chain scenes toward a minute or more. • Insert: add an object, with Flow adjusting shadows and scene lighting so it sits naturally. • Flow Tools and Flow Agent (May 2026): Tools are reusable presets you create in plain language, such as a resize or a look. The Agent is a Gemini-based assistant that helps from idea to dialogue to edits. • Gemini Omni Flash: supports conversational editing, where you ask for a change to a character or background and the rest of the video holds steady. Google says Omni videos carry SynthID watermarking. On cost, a third-party tracker updated October 20, 2026 reports that Flow’s free tier gives 50 daily credits on Veo 3.1, with Fast generations costing roughly 20 credits and Quality roughly 100. Credit schemes change often, so confirm in Flow’s own plan page before you budget. If you are searching for AI shortcuts to use free, this is the legitimate route: spend free credits on short, well-briefed clips, not on experiments. Kling 4.0: Direct Through Keyframes and References Kling’s pitch for version 4.0 is control rather than raw quality. Reported specs include prompts up to 8,000 tokens, 30-second native clips, up to 10 keyframes and up to 15 combined references, with stereo audio and editing tools that change people, backgrounds or weather in existing footage. Kuaishou also lists 4K and 10-bit HDR output, but several outlets describe those as still “coming soon,” so check the current plan page. Camera language is where Kling has been getting attention. Coverage describes understanding of push-ins, pull-outs, pans, tracking and orbits. In one independent same-prompt test of five models, Kling was the only one whose camera orbited the subject, but that was a single seed, so treat it as a hint, not a verdict. A useful pattern is a timed keyframe script: Keyframe 1 (0s): wide establishing shot, empty night market, rain on awnings. Keyframe 2 (8s): medium shot, vendor lifts the lid of a steaming pot, camera tracks left. Keyframe 3 (18s): close-up on a child’s hands taking a paper bowl, shallow focus. Keyframe 4 (28s): slow pull-out to wide, lanterns glowing, crowd murmur fading. Longer prompts do not mean better prompts. Use the extra space for structure, not adjectives. Midjourney: The One Tool With a Real Command Grammar Midjourney is the clearest case of a true command modifier. Parameters go at the end of the prompt, start with two dashes, take a space before their value and use no punctuation. Text placed after a parameter is ignored or breaks the prompt. Parameter Job Notes --ar 9:16 Aspect ratio Set it before animating --stylize or --s Strength of Midjourney’s own aesthetic 0 to 1000, default 100 --style raw Less default styling Better for literal control --sref and --sw Style reference and its weight Reuse a look across a series --seed Repeatable starting noise Good for A/B tests --no Excludes elements Keep the list short --chaos and --weird Variation and oddness Use for exploration --motion low or --motion high Video movement Low is the default --loop and --end Looping and end-frame control For video jobs --bs Number of videos per prompt Batch size Version warning: parameter support changes. Midjourney’s documentation now describes --oref as replaced by the Edit Model in V8, and third-party guides disagree on which flags V8.1 and V8.2 still accept. Run a quick test on your version before you build a template around any flag. A still-image example: weathered fisherman mending a net on a wooden pier at blue hour, 50mm lens feel, shallow depth of field, rim light from a sodium street lamp, fine film grain --ar 3:2 --style raw --stylize 100 --no text Then animate it with --motion low for a subtle push-in, or --motion high when you want big camera moves and can tolerate glitches. Six Prompts You Can Use Today 1. Product teaser (Veo in Flow, text to video) Smooth 360-degree orbit around a matte black wireless speaker on a pale stone pedestal, softbox key light from camera left with a thin rim light behind, shallow depth of field, minimal studio background, muted grey and warm sand palette, soft ambient room tone, 8 seconds, 9:16. 2. Image-to-video motion note Keep the framing and light exactly as in the image. Slow dolly-in on the chef’s face. Steam drifts upward from the pan. Quiet kitchen ambience and a low sizzle. No cuts, no added people. 3. Documentary handheld Handheld medium shot following a street vendor through a crowded morning market, natural overcast light, slight camera shake, shallow focus on the vendor’s shoulder, muffled chatter and a bicycle bell. 4. Dramatic reveal Low-angle shot at dusk, an astronaut stands on a ridge, camera cranes up to reveal a vast red valley, backlit by a low sun with long shadows, dust drifting through the beams, a quiet wind and a faint radio crackle. 5. Locked-off interview look Locked-off tripod shot, no camera movement, medium close-up of a woman at a desk, soft window light from camera left, subtle fill from the right, 85mm feel, blurred bookshelves, quiet room tone. 6. Style-matched series (Midjourney) lighthouse keeper’s kitchen, morning light through a small window, painterly, muted greens --ar 16:9 --sref [your reference image URL] --sw 100 Swap the nouns, keep the structure. That is what makes a prompt library reusable. Claude as Your Prompt Director Claude does not generate video, but it is an excellent shot-brief writer, because the job is structured language. Anthropic’s documentation recommends clear and direct instructions, three to five well-built examples wrapped in tags, and XML tags to separate instructions, context and input. That maps neatly onto video work. You are a commercial cinematographer briefing an AI video model. Turn the idea below into four shots of six seconds each. - One camera move and one lighting setup per shot. - Describe only what the camera can see and hear. - Repeat the subject description word for word in every shot. A quiet teaser for a handmade ceramic tea set, 9:16, premium and calm. For each shot: one paragraph of 60 to 90 words, then a line starting CAMERA: Paste your favourite example into tags once and Claude will follow the pattern for every project. Are There Secret Claude Prompt Shortcuts? Not in the sense the internet means. There is no hidden phrase that flips a switch. What works is plain craft: • State the purpose and audience. “For a 15-second Instagram ad aimed at first-time buyers” beats “make it good.” • Show an example. Models generalise strongly from samples, so if every example is short, the output will be short. • Use tags. , and keep long prompts unambiguous. • Ask for revision passes. “Draft, critique against my rules, then rewrite” is a legitimate shortcut. • Save the repeats. In claude.ai, Projects and custom styles hold your standing instructions. In Claude Code, skills do the same job (next section). Claude Code Slash Commands and Skills This is the one place the phrase “custom slash commands” is literal. Claude Code’s documentation says custom slash commands have been merged into skills: a markdown file with instructions, plus optional YAML frontmatter, that becomes a /command. Existing files under .claude/commands/ keep working, but new automation should use skills, which add supporting files, automatic invocation and invocation controls. Where they live • Personal skills: ~/.claude/skills//SKILL.md, available in every project. • Project skills: .claude/skills//SKILL.md, committed with the repo so a team shares them. • Plugin skills: inside an installed plugin. A working example you can adapt for video planning --- name: shotlist description: Turn a one-line video idea into a numbered shot list with camera, light and audio notes. argument-hint: [video idea] disable-model-invocation: true --- Create a shot list for: $ARGUMENTS Rules: 1. Four to six shots, five to eight seconds each. 2. One camera move and one lighting setup per shot. 3. Repeat the subject description word for word in every shot. 4. End each shot with an AUDIO line. Output a table: Shot | Prompt (60-90 words) | Camera | Light | Audio. Save that as ~/.claude/skills/shotlist/SKILL.md, then type /shotlist a calm teaser for a ceramic tea set. The $ARGUMENTS placeholder receives whatever you type after the command. If you leave the placeholder out, Claude Code appends your input to the skill content so nothing is lost. Setting disable-model-invocation: true turns the skill into a pure slash command that only runs when you call it, which is the right choice for anything with side effects. Keep each SKILL.md focused. Community guidance suggests staying under about 500 lines and moving heavy reference material into separate files. Ollama Commands: A Free Local Prompt Lab Ollama runs open models on your own machine, which makes it a cheap sandbox for drafting and rewriting prompts. It will not render video, but it can pre-process your shot lists at zero per-token cost. Command What it does ollama pull Downloads a model ollama run Starts an interactive chat ollama list Shows installed models ollama ps Shows running models ollama stop Unloads a running model ollama show Displays model details ollama create -f Modelfile Builds a custom variant ollama serve Starts the server Inside an ollama run session, slash commands are real: /show info, /show system, /show modelfile, /set parameter and /bye. To build your own shot-director model, create a file named Modelfile: FROM llama3.2 SYSTEM You are a cinematographer. Convert any idea into one video prompt with subject, action, camera, light and audio. PARAMETER temperature 0.7 Then run ollama create shot-director -f Modelfile followed by ollama run shot-director. Ollama also speaks the Anthropic Messages API. Its documentation shows ollama launch claude configuring Claude Code to use a local or cloud model, or you can set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN yourself. Expect gaps: Ollama says the API key is not validated, and some Anthropic features, such as the Batches API, are not supported. Developers also report mixed results with Claude Code on small local models, so test before you rely on it. How to “Trick” Claude to Do Anything You cannot, and the framing is the wrong way to think about it. Anthropic designs Claude to decline a small set of harmful requests regardless of phrasing, so clever wording is not a durable skill and I will not publish a jailbreak recipe here. What people usually experience as “Claude refused” is much more mundane: the request was ambiguous, missing its purpose, or too big. The fixes are legitimate and effective. • Add the why. “I am a marketer writing a fictional villain monologue for a game trailer” gives Claude the context that a bare request lacks. • Split the task. Ask for the outline, then each section. • Name the format. Table, script, JSON, bullet list. • Ask what is missing. “Before you answer, list any assumptions you are making.” If you want Claude to do more, give it better inputs. Tricks expire. Clarity does not. A Prompt for Claude to Write Like a Human No prompt makes text undetectable, and chasing that goal produces worse writing. A better target is writing that sounds like a specific person with a specific job. Three moves matter most: 1. Show it your voice. In claude.ai, open the style menu, choose Create and edit styles, then upload a writing sample or describe the style. Claude analyses the sample and saves a reusable custom style. 2. Describe what you want, not only what to avoid. Anthropic’s guide demonstrates steering format by telling Claude what to produce, such as clear, flowing prose in complete paragraphs. 3. Demand specifics. Generic text comes from generic inputs. Feed it real numbers, names and a viewpoint. A prompt that works: Write as a working marketing consultant explaining this to a smart friend. Use plain words, varied sentence length, and one concrete example per point. Include a small opinion where I would normally hedge. Match the rhythm of the sample below. After drafting, cut any sentence that could appear in any article about any topic. [paste 300-500 words of your own writing] [your topic and the facts to include] That last instruction, the “could appear anywhere” cut, removes most of what readers recognise as machine-written filler. Myths vs Reality Myth Reality There are secret slash commands for Veo, Flow or Kling No. They respond to descriptive prompts, references and keyframes Longer prompts always win Structure wins. Kling accepts very long prompts, but one move per shot still matters Sora is the best option for cinematic prompts Sora is shut down. Use the principles with Veo, Kling or others Midjourney parameters are the same across versions They are not. Test flags on your version Claude Code commands and skills are separate systems They are unified. Either file format creates the command Local models match Claude for agent work Sometimes for simple tasks, but results vary widely Your Prompt Engineering Course Path If you want structure, start with the free material from the source. Anthropic publishes an interactive prompt engineering tutorial that moves from basic prompt structure to reusable, complex prompts, and its Anthropic Academy hosts free self-paced courses, including Claude Code in Action for terminal-based workflows. For video craft, Google’s Veo prompt guide on Vertex AI is the best primary reference on prompt anatomy, camera angles and movements. Practise by taking one idea and building the same six-shot sequence in two different tools. The differences teach you more than any listicle. Frequently Asked Questions How do I use AI to generate videos? Plan the shots, create still keyframes, animate them with image to video, then extend and edit. Use the six-layer brief for every prompt. What are effective prompts for AI video generation? Prompts that name one subject, one action, one camera move, one light setup and an audio cue. The six examples above are templates. Are there free AI shortcuts for video or image generation? Flow offers a free credit tier, which a tracker pegs at 50 daily credits as of October 2026. Ollama and Claude Code skills help you plan for free or at low cost, but verify limits on each plan page. What are secret Claude prompt shortcuts? None are secret. Tags, examples, styles, Projects and Claude Code skills are the real ones. Can I make Claude do anything? No. Give clearer context and better inputs instead. Does Google Flow have commands? Not slash commands. It has Frames to Video, Ingredients to Video, Extend, Insert, Flow Tools and Flow Agent. Conclusion: Build a System, Not a Spell Book The most useful “command” you can learn is a habit: say who is in the shot, what they do, where the camera sits, how it moves, where the light comes from, and what we hear. Everything else is packaging. To start today: 1. Copy the six-layer template and brief your next shot with it. 2. Generate a still first, then animate it. 3. Save the template as a Claude Code skill so /shotlist is one keystroke away. 4. Verify every tool claim against the vendor’s page before you invest, because this market now changes monthly. A tool that was central in January can be gone by April, as Sora showed. SEO Package • Suggested title tag (under 60 characters): AI Video Prompt Workflow: Camera, Light and Commands • Meta description (under 155 characters): Learn the master prompt workflow for AI video and images: camera angles, lighting, Google Flow, Kling 4.0, Midjourney and Claude Code commands. • Suggested slug: ai-shortcuts-slash-commands-video-prompt-workflow • Secondary keywords woven in: claude code skills, claude code custom slash commands, google flow ai, google flow commands, image to video, ollama commands, prompt engineering course Sources and Verification Notes Primary and vendor sources: Claude Code documentation (code.claude.com/docs/en/slash-commands); Anthropic prompting best practices (platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices); Anthropic support article on styles (support.claude.com/en/articles/10181068); Google blogs on Veo 3.1 in Flow (blog.google/innovation-and-ai/products/veo-updates-flow/) and Ingredients to Video (blog.google/innovation-and-ai/technology/ai/veo-3-1-ingredients-to-video/); Google Cloud Veo prompt guide (docs.cloud.google.com/vertex-ai/generative-ai/docs/video/video-gen-prompt-guide); Midjourney parameter list (docs.midjourney.com); Ollama Anthropic compatibility docs (docs.ollama.com/api/anthropic-compatibility). News and secondary sources: 9to5Google and Decrypt on Flow and Gemini Omni (May 2026); BigGo Finance, UnificAlly and SandBase on Kling 4.0 (October 2026); API Frame, Tech Insider and PYMNTS on the Sora shutdown; CostGoat for Flow credit pricing; fal.ai’s summary of OpenAI’s Sora prompting guide. Not independently confirmed: Kling 4.0 HDR and 4K availability (sources conflict), Flow credit prices (third-party tracker), Midjourney V8.1 and V8.2 flag support (sources disagree), and any Sora revival. Re-check these before publishing.](https://tjpromotions.pro/storage/2026/10/Learn-the-master-prompt-workflow-for-AI-video-and-images-camera-angles-lighting-Google-Flow-Kling-4.0-Midjourney-and-Claude-Code-commands-1290x725.png "AI Shortcuts, Slash Commands and Camera Control: The Master Prompt Workflow for AI Video and Images (2026) 1")
